What Is
Commercial Real Estate
AI?

The short answer: commercial real estate AI is software that uses machine-learning models to score, predict, or recommend actions on commercial properties — combining public data, proprietary data, and pattern-recognition algorithms to surface signals that would take a human analyst weeks to compile.

The longer answer is what everyone selling it is trying to confuse you about.

I'm writing this as someone who does commercial real estate for a living in Oklahoma City and who spent the last two years building an AI platform for the work I do. I'm not a software company trying to sell you a seat. I'm a broker who got tired of spending Sunday nights in spreadsheets and decided that was a solvable problem.

So when people ask me, what is commercial real estate AI, really? — here's the plain-English version.

Commercial Real Estate AI, Defined

Commercial real estate AI is not a chatbot on a brokerage website. It's not CoStar with ChatGPT bolted on. It's not a startup pitch deck.

At its core, CRE AI is three layers of software working together:

1. A data foundation — structured information about properties, transactions, demographics, permits, ownership, zoning, and dozens of other signals.
2. A modeling layer — statistical or machine-learning models that find patterns in the data and produce predictions, scores, or rankings.
3. An interface — a way for a human to ask questions and see answers, usually a map, dashboard, or natural-language search.

That's it. Everything else is marketing.

The reason people get confused is that the word "AI" is doing a lot of work right now. In 2026 it covers everything from a linear regression that's been running for 15 years to a large language model that hallucinates cap rates. A real CRE AI platform blends both: deterministic models for the stuff you need to be right about, and generative models for the stuff you need to summarize.

What makes a platform worth paying for is how well those four stages are actually built. Most of what gets marketed as "AI for commercial real estate" leans heavily on stage one (lots of data), glosses over stage two (no real modeling), paints stage three as the product (a score without a methodology), and punts on stage four (no workflow a working broker would actually use).

A good CRE AI platform is honest about all four.

What CRE AI Can — And Can't — Do

I get asked this in every conversation. Here's the honest version.

What it can do (right now, in 2026)

• Parcel scoring. Rank every commercial parcel in a market by investment potential, risk, or development pressure. Signal Intelligence scores 504,000+ parcels across Oklahoma County.
• Price prediction. Estimate values with 5–10% accuracy at the market level, using hedonic regression or gradient-boosted models trained on decades of transactions.
• Off-market discovery. Flag properties likely to come to market soon, based on ownership tenure, loan maturities, distress signals, or permit activity.
• Feasibility acceleration. Run a preliminary site analysis — demographics, traffic, competitor mapping, zoning overlays, submarket comparables — in seconds instead of days.
• Pattern detection. Surface cluster patterns a human would miss: the three submarkets absorbing the same tenant type, the six parcels with correlated permit activity, the warehouse owners with similar hold periods and LTVs.

What it can't do (yet, and maybe never)

• Replace a broker. It doesn't walk the site. It doesn't know the seller had a second divorce. It doesn't know the city is about to zone the adjacent block. The market is physical, emotional, and political. AI is not.
• Predict black-swan events. Pandemic-era office collapse. 2008. Any rapid Fed pivot. Historical data trains AI on things that have happened, not things that will surprise it.
• Work without good data. Garbage in, garbage out is louder in CRE than in almost any other field, because CRE data is unusually fragmented, proprietary, and regional.
• Hallucinate your way to a deal. An LLM that confidently invents a cap rate is worse than no AI at all. Real CRE AI is deterministic where it matters, generative where it doesn't.

The word “AI” is doing a lot of work right now. It covers everything from a linear regression that's been running for 15 years to a large language model that hallucinates cap rates. — Aaron Diehl

The Four Kinds of CRE AI Builders

Not all commercial real estate AI is built by the same kind of company. In 2026, the landscape breaks into four rough camps — and knowing which camp built a given tool tells you most of what you need to know about whether it'll serve you well.

There's nothing wrong with any of these categories. The enterprise platforms do work an independent broker never could. The data vendors have decades of information no startup can replicate. The specialists often outperform the generalists on the narrow thing they do.

But none of them are built by someone who spent last Wednesday arguing with a landlord about a tenant improvement allowance.

Why Broker-Built Matters

This is where I have a point of view.

I've used every category above. Some of them are genuinely good. But every one of them was built by software engineers who've never done a deal, and it shows in the product. The questions the software is optimized to answer aren't quite the questions a working broker actually asks.

A software engineer asks: what can we do with this data?

A broker asks: which parcels should I call on this week?

Those are different questions. They lead to different products. And until very recently, almost every CRE AI product was answering the first one.

The broker questions AI should be answering

• Which owners in my territory are most likely to sell in the next 12 months?
• Of the 30 properties that fit my buyer's box, which three have the best gravity profile?
• Where is tenant demand concentrating that the market hasn't priced in yet?
• Which submarket's absorbing rent growth fastest, and which is about to?
• What's a fair price for this parcel given its specific mix of characteristics — not the median comp from three miles away?
• Where are the off-market deals hiding?

Those are the questions Signal Intelligence was built to answer. Not because they're technically interesting — they often aren't — but because they're the questions I actually have to answer, every week, to do my job.

That's the bet. That the best CRE AI isn't necessarily the one with the most data or the most impressive-looking model. It's the one that answers the questions that actually move deals.

So, Is It Worth It?

If you're a broker, developer, investor, or lender in commercial real estate, and you have more research to do than time to do it, then yes — some form of CRE AI is now an expected part of the stack. Not optional. Not cutting-edge. Expected.

The real question isn't whether to use CRE AI. It's which kind, for which question, in which market.

A national firm needs enterprise-grade coverage. A specialist fund needs deep data and modeling they can stress-test. A working broker in a specific market needs something that speaks the language of that market — the submarkets, the owners, the recurring patterns — and answers the week-to-week questions that actually close deals.

That's what Signal Intelligence is for.

If you want to see the product itself, explore the Gravity Map. If you want to understand how the scoring actually works — how gravity waves and convergence produce a signal, with live animations — the next essay in the series goes deeper there. And if you want to see 25 years of how Oklahoma County CRE has performed against the national market, the Diehl Index is live.

And if you want to be one of the first people to use it, there's a waitlist below.